## Bitget Wallet AI Agent Direction
Don't fight the wrong war.
AI is not a reason to build a wallet for machines. It is the chance to turn a self-custodial dollar account into a product ordinary people can use.
### Topic: AI × Bitget Wallet
North Star: The AI dollar account for emerging markets
Date: 2026.06 · fact-checked
### THE THESIS
Do not turn the AI strategy into an agentic wallet or a generic chatbot. The real job is to use AI to fold funding, FX, transfers, getting paid, yield, safety and support into one intent entry you complete by saying a sentence in your own language, with the wallet running invisibly underneath. First, help emerging-market users complete three actions: first funding, first transfer, first PayFi use. Then use user-authorized behavior data for risk, underwriting and yield recommendations. AI's value is not showing off. It is turning self-custody from a power-user product into a mainstream finance entry point.
### 01 First, be clear what AI actually changes about money
Don't start from "should we add an AI feature." Start from first principles: the three things AI structurally changes about money.
#### THREAT: Interface collapse
The app and the screen are no longer the entry point; intent is. This threatens every app, including ours. Whoever owns the intent layer owns the user.
#### 01 OPPORTUNITY: A new economic actor
Autonomous agents transact, hold and pay, but can't use card or bank rails (no human identity, micropayments, 24/7). Crypto wins the long tail, but it's still tiny today.
#### 02 OPPORTUNITY: Complexity goes to zero
Bridging, swaps, gas, yield, risk: AI does it all. This removes crypto's single biggest adoption barrier, the one at the heart of our emerging-market strategy.
Remember these three. AI is simultaneously our biggest entry-point threat (#1) and our biggest product opportunity (#3). The one everyone is crowding into, #2, is worth the least right now.
### 02 Reality check: what not to misread in June 2026
Three judgments can easily pull resources in the wrong direction: agentic wallets, payment protocols and agent-payment scale. They are not forbidden, but they are not the main battlefield.
✕ Misread A: "shipping an agentic wallet creates a lead"
Do not rush. The problem is not whether we are late; it is that the category still lacks a clear real user and a complete use case. Coinbase, OKX, Trust Wallet, Binance, and TON all shipped or promoted related capabilities in a short period, but most are still capability demos. The rule is simple: first explain who uses it daily and why they stay or pay, then ship. If we cannot explain that, keep watching.
✕ Misread B: "we should build our own protocol to rival x402"
Wrong. This is a standards fight, and the winner is chosen by developers, cloud platforms, payment networks and merchants together. x402 has joined the Linux Foundation and gathered support from Google, Cloudflare, Stripe, AWS; OKX APP has brought in AWS, Alibaba Cloud, Ethereum, Solana, Uniswap and others. We do not need to define a global standard. The right move is to adopt the winner, not build a protocol nobody uses.
✕ Misread C: "agent payments are already a large market"
Half wrong. Keyrock reported that from May 2025 to April 2026, AI agents completed about 176 million on-chain payments totaling over $73 million, 98.6% in USDC, with 76% below $0.30. This proves the direction is real, but still small. It is a Morph option and standards-support bet, not the wallet's main battlefield.
OpenAI deprioritized native checkout; shifted to discovery and merchant checkout.
$73 million in 176 million agent on-chain payments; 98.6% USDC. $33 trillion stablecoin volume; 85-90% is trading. Stripe bought Bridge / Privy; entered the AWS AgentCore payments stack.
The point to notice: OpenAI's case shows that ChatGPT can shape product discovery, but checkout still falls back to merchants and Stripe/card rails. Consumer e-commerce checkout is not decided by "LLMs choosing stablecoins"; it is decided by merchants, risk controls, refunds, disputes and local payment methods. Stablecoins break through more naturally in cross-border, high-frequency low-value and machine-to-machine payments, not ordinary e-commerce checkout.
### 03 The real prize is in people, not machines
The "dollar-account" behavior in emerging markets already formed on its own; nobody had to teach it. What's missing isn't demand, it's product.
- 50%+ of Turkey's crypto transactions are stablecoins (4.3% of GDP) - 61.8% of Argentina's volume is stablecoins - 26 million Nigeria stablecoin users, 12% penetration - +80% South Asia on-chain YoY; India #1 globally
The behavior is validated. AI can remove the last 20% of friction: local on/off-ramp, bills, merchant QR, unfamiliar terms and safety judgment. But do not mistake a simple interface for a long-term defense: simplicity is the entry ticket, not the reason users stay.
The real value comes after: once people are in, user-authorized behavior (how they fund, how much they hold, when they convert, where the PayFi card is spent) becomes a signal others do not have. Use it for underwriting, risk and yield recommendations, plug it into PayFi credit already available in 60+ countries, and the loop strengthens with use:
bring people in → user-authorized signals → underwriting / risk / yield → PayFi revenue → lower acquisition cost → more people
This loop answers two questions: why users stay, and where revenue comes from. One hard constraint: the loop must work under self-custody. Signals come only from what we can see and what the user authorizes (PayFi card spend, on/off-ramp flows, user-authorized data), not from holding users' assets. So it does not conflict with "only you can move your money."
Around that loop sits an operational barrier: regulatory relationships, local rails and trust. Even with a 3 billion-user entry point, Meta's WhatsApp Pay flopped in India, beaten by local Google Pay and PhonePe. US platforms usually move years slower in heavily regulated emerging markets. This is a window, not permanent safety: miss it and it closes.
### 04 What AI actually does in the product
This section turns the direction into product. Judge every AI feature by three tests: does it improve first funding, first transfer and first PayFi use; does it reduce support and risk cost; does it create user-authorized signals we can use. If not, it is not a core AI project. AI support and market copilot are already live; the rest should be validated by priority.
#### P1 PILOT · INTENT SEARCH
One search box that gets what you mean.
Natural language reads intent and routes it: funding, FX, transfers, tokens, US stocks, support and features. "Buy some Apple" goes to US stocks; "cheapest USDT transfer" goes to send; "how do I withdraw" goes to help.
Validation: Measure whether new users complete first funding, transfer and PayFi use faster.
#### P1 PILOT · NEW-USER JOURNEY
The avatar and minus-one screen, guiding the next step.
Use two surfaces, the top-left avatar and the minus-one (left-swipe) screen: AI reads the user's current intent and state, then proactively guides and recommends what to do next. The clearest case is walking a new user seamlessly from sign-up to a first transaction.
Validation: Measure KYC, funding, first transaction and first PayFi conversion step by step. Do not hide behind one blended conversion rate.
#### P1 PILOT · VOICE INPUT
Talk to it, not only for support.
